Grok 4.1 Fast (Reasoning) by xAI

Key takeaways

• Grok 4.1 Fast (Reasoning) is an agentic tool-calling model by xAI, designed for real-world applications like customer support and deep research, featuring a 2M context window and multimodal input (text+image).
• It utilizes a proprietary Mixture of Experts (MoE) architecture with 300 parameters, but its weights are not open-source, and it generates text outputs.
• The model excels in general knowledge and reasoning across undergraduate subjects (91.09% accuracy on MMLU) and strong in mathematical reasoning (70.59% accuracy on AIME 2025).
• A significant limitation is its complete failure (0% accuracy) on software bug-fixing tasks on SWE-bench Lite, attributable to an inability to manage dynamic execution environments and generate correct, executable code changes.
• While proficient in foundational math and core science, it struggles with advanced geometry, complex number theory, combinatorial problems, and practical code execution, often misinterpreting complex logical statements and lacking contextual awareness for tool usage.

Technical specifications

Context length

2M tokens

Max output

—

Architecture

MoE

Modalities

text+image->text

Parameters

~300B

Open weights

No

Licence

Proprietary

Regions

USA

Vendor

xAI

Provider model key

x-ai/grok-4-1-fast-reasoning